The issue of water availability has become crucial, particularly during the summer months when droughts are becoming increasingly frequent. Anticipating water levels in the water tables is a major challenge for the sustainable management of water resources.
A major challenge for society and the environment
It is in this context that ILLUIN Technology and BRGM (Bureau de Recherches Géologiques et Minières) have collaborated to develop an innovative approach based on Machine Learning, to predict the evolution of water tables and better plan their management.
Predicting water levels in aquifers is essential to avoid water shortages and optimize water use, particularly in regions like the Adour-Garonne basin that face prolonged periods of drought. BRGM wanted to explore the possibilities offered by Machine Learning to improve the accuracy of predictions based on historical aquifer data.
ILLUIN Technology rose to the challenge, contributing its expertise in Data Science to create predictive models capable of anticipating short- and medium-term fluctuations in water levels.
A Machine Learning approach to predicting water levels
To meet BRGM's needs, the ILLUIN Technology team, comprising Yonatan Deloro, Michaël Resplandy and Théo Rubenach, set up several statistical regression models. These models were tested on several aquifers in the Adour-Garonne basin, with the aim of forecasting water levels over a period of 3 to 6 months.
The innovation of this approach lies in its flexibility. Two types of model were tested:
-
-
- Local model: A model specific to each aquifer, which learns from historical data for that monitoring point only.
- Global model: A model based on data from several aquifers sharing similar hydrogeological characteristics, useful when data history is insufficient.
-
Promising results: accurate prediction despite limited data
The conclusions of this project are particularly encouraging:
-
-
- Local model: When the data history is sufficient (several years), the local model offers highly accurate results, adapting to the specific characteristics of the aquifer.
- Global model: In the absence of a long data history, the global model has proved extremely effective. It reduces the mean error (RMSE) by up to 75% compared to a local model based on only two years of historical data.
-
This approach makes it possible to obtain reliable predictions even for aquifers with little data, a decisive advance for groundwater management.
Next: Compare and refine models
The next stage of this project is to compare the performance of the models developed with the traditional approaches used by BRGM. The aim is to validate that these Machine Learning -based models can deliver faster, more accurate predictions, while requiring less modeling work by hydrogeological experts.
One of the great advantages of this method is that it doesn't require any specific modeling effort on the part of business experts, making it easy to adopt and use on a large scale.
Conclusion: A technological breakthrough for water resource management
By collaborating with BRGM, ILLUIN Technology is actively contributing to the ecological transition by offering innovative technological solutions for the sustainable management of water resources. Thanks to Machine Learning, it is now possible to predict groundwater levels more accurately, even when available data is limited.
This advance not only optimizes water resource management, but also enables us to better anticipate periods of drought, thereby reducing the associated environmental and economic risks.
